HPC Engineer
About this role
At Verda, we're building a full-stack AI cloud, covering everything from data centers and hardware to our own cloud platform that the world's leading AI teams use to do serious AI work.
We strive to make a positive mark on the world through the infrastructure we build and give leading teams a service they can truly depend on. Headquartered in Helsinki, we operate globally with offices in London and San Francisco.
Join Verda while it’s still being built - not once it’s finished.
Why Verda
Cash and equity compensation along with various fringe benefits.
Profitable operations with rapid, sustained growth.
30+ nationalities, with 6 different ones on the management team.
A real chance to make an impact and work alongside world class engineers, researchers, and partners across the global AI ecosystem.
Practicalities
Work mode: Remote (EU)
Level: Mid / Senior
Employment type: Full time and permanent
About the role
GPUs only deliver value once they're wired into a cluster that researchers can actually train on. As our HPC Engineer, you'll own the baremetal and virtualized clusters behind our AI cloud - from the InfiniBand fabric and shared filesystems up through the workload orchestration layer. You'll be the person who keeps large-scale GPU clusters healthy, performant, and ready for the next workload.
Your responsibilities
Administer baremetal GPU/HPC clusters end to end, from provisioning through day-two operations
Administer virtualized clusters, including the hypervisor and GPU virtualization stack underneath them
Design, deploy, and tune InfiniBand fabrics, including topology planning, subnet management, and performance validation
Deploy and operate shared/parallel filesystems supporting training and inference workloads, balancing performance, capacity, and reliability
Troubleshoot and resolve issues across the full fabric stack: fibers, transceivers, NICs, switches, drivers, and firmware
Partner with remote-hands and data center teams to diagnose hardware faults and execute physical-layer fixes and cluster expansions
Operate and tune Slurm (or equivalent) workload scheduling deployments used by customers and internal teams
Keep issue tracking, IPAM, and DCIM records accurate as clusters are built, changed, and decommissioned
Participate in on-call rotations and incident response for cluster-level issues
Collaborate with platform, network, and storage teams to integrate new clusters into the broader AI cloud
Your key competencies
Solid Linux skills, with specialization in memory management, PCIE topologies and virtualization being a bonus
Deep Infiniband knowledge, including fabric design, subnet management, and performance tuning
Solid experience with IB clustering, troubleshooting/debugging, understanding the ecosystem of fibers+transceivers+NICs+switches and all the things that could possibly fail in them
Experience about shared filesystems (e.g. Lustre, GPFS/Spectrum Scale, WekaFS, or similar)
Ability to work with remote hands teams to diagnose and resolve hardware issues remotely
Knowledge about NCCL, CUDA, DOCA and the Nvidia stack
Knowledge about Slurm and/or other workload scheduling solutions, bonus points for Slinky/slurm-bridge
Understanding the importance of keeping issue tracking/IPAM/DCIM up to date
Comfort operating production clusters where uptime and performance directly affect customer workloads
Scripting/automation ability (e.g. Python, Bash, Ansible) for repeatable cluster operations
Nice to have
RoCEv2 knowledge (and/or Spectrum-X)
Understanding of agentic guardrails, especially when applied to administration of complex systems
Ability to think beyond what is needed right now vs. some given trajectory or roadmap
Experience with GPU health-checking and diagnostics tooling (e.g. DCGM, field diagnostics)
Experience with baremetal provisioning/orchestration tooling (e.g. MAAS, Foreman, custom PXE/iPXE pipelines)
Familiarity with GPU-aware virtualization or containerization (Kubernetes device plugins, KVM/QEMU with GPU passthrough, SR-IOV)
Exposure to observability stacks (Prometheus, Grafana, Loki) for cluster-level monitoring
What's next
We're building fast and this role needs the right person behind it. There's no artificial deadline, but when we find who we're looking for, we move. If this sounds like your next move, apply now.
Please submit your application through our Careers page. We don't accept applications sent by email.
Originally posted on Himalayas
